A new creep model directly using tabulated test data and implemented in ANSYS

XIV International Conference on Computational Plasticity. Fundamentals and Applications

COMPLAS 2017
Kang Shen, Wilhelm J.H. Rust

A NEW CREEP MODEL DIRECTLY USING TABULATED TEST DATA

AND IMPLEMENTED IN ANSYS
KANG SHEN, WILHELM J.H. RUST

Hochschule Hannover –University of Applied Sciences und Arts

Faculty II department of Mechanical Engineering
Ricklinger Stadtweg 120 30459 Hannover, Germany
E-mail: kang.shen@hs-hannover.de, wilhelm.rust@hs-hannover.de

Keywords: creep, material model, tabulated test data, ANSYS.

Abstract. Nowadays plastics are increasingly used in highly stressed structures in all kinds of
constructions. The time dependency, the so-called viscosity, is a crucial part of the material
behavior of plastics. A typical form of viscosity is creep. Creep is the increase of deformation
under constant load. In the FE-simulation creep behavior is usually described by creep law
functions. The commercial software provide many creep law functions depending on time,
stress, strain, temperature and multiple material parameters. To run a creep simulation, the user
must define all the parameters which requires a certain effort. Curve-fitting procedures might
be of help, the results, however, often are not precise enough. For these reasons, we introduce
our new creep model doing the similar job as the creep law functions but being able to directly
use the tabulated data of the creep tests without curve-fitting procedures. In this paper, we use
the model to create a 3D stress-creep strain-time surface based on the tabulated data like isoch-
ronous curves, which is represented by bicubically blended Coons patches to provide a good
convergence due to their differentiability. This creep model supports strain hardening, which
shows more realistic behavior when the load changes significantly during the simulated process.

1 INTRODUCTION
Numerous plastics show a combined material behavior like elastic, plastic and viscous one.
The viscosity of plastics strongly depends on the loading rate, the loading duration and the
temperature [1, 2]. Ignoring the viscosity of plastics like creep can lead to a severe failure of
construction based on simulation [3]. Creep describes the increase of the deformation with time
under a mechanical stress [4].
The simulation of creep behavior is based on the creep test data. The creep is directly meas-
ured from the creep and relaxation test. Creep curve provides a dependence of creep strain on
time, relaxation curve provides a dependence of stress on time while creep strain can indirectly
be determined. The isochronous curves are concluded from those two types of curve [5]. The
isochronous curve gives a stress-strain relation at the identical time point.
Another well-established curve for creep is the creep modulus curve. The creep modulus is
the quotient of the stress and total strain, thus showing a combination of creep and elastic be-
havior over time. Considering Young’s modulus as constant there is a direct relation to creep
strain.

319
 Kang Shen, Wilhelm J.H. Rust

Besides of stress and time creep behavior of plastics also depends on temperature. Conven-
tional creep models usually describe the temperature dependency using the Arrhenius equation
showing exponentially growing creep rates with respect to the inverse of the temperature [6].
Modelling temperature dependency in this way shows limited accuracy. On the other hand creep
rate indeed varies nonlinearly with the temperature. Thus, the alternative, linearly interpolating
all creep parameters might better match the material behavior for the given temperatures but
lacks in between.

2 MOTIVATION
Numerous creep models proposed creep law functions defining creep behavior. For those
functions several more or less abstract, not directly measurable material parameters must be
specified. The difficulty is to determine the parameters as accurate for general application. Of-
ten one creep curve for a single stress can be fitted accurately. The other dependencies, those
of temperature and stress, then show larger errors if - which is often the case - the curves for
different stress levels are not similar in the mathematical sense, i.e. not scalable. Furthermore,
it has to be emphasized that usual creep models give functions not for creep strain but for strain
rate. Curve fitting requires either the solution of the differential equation given in this way or
the determination of strain rates from measured strain which can be difficult due to either a
small number of sampling points or oscillations in the measured curves if the number of sam-
pling points is larger. The study of [7] indicate that general curve fitting functions could create
insufficiently accurate or even wrong parameters and mislead the simulation. As a consequence
the user has a larger effort to determine creep parameters and will even though finally not be
satisfied.
Furthermore, due to numeric (time integration scheme) creep rate models can be subject to
larger initial errors, e.g. for the power function of time with negative exponent, even if the
analytical use of the parameters show good accordance with tests. The latter is the reason why
the authors avoid rate formulations but use incremental ones as shown below.
Despite the right selection of creep model and define its parameter, the material behavior of
a creep model can also differ from the test data[8,9]. The reason for that is the fact that constant
material parameters compromise the accuracy with multiple loads and temperatures. With the
compromise the material behavior in simulation is only accurate in a limited range [10].
In this work, we focus our attention on the construction of a creep model with direct use of
tabulated test data from the creep test. The tabulated data here are e.g. the creep-time curve,
isochronous curve and creep modulus curve. This creep model works without parameter iden-
tification and the curve fitting function. Instead of the parameters it uses a three-dimensional
surface to find the proper creep state for calculation. For a good differentiability and a decent
extensionality, we decided to use bicubical Coons patches (also named bicubically blended
Coons patches) for the surface construction.

3 CONVENTIONAL CREEP MODEL

Numerous conventional creep models describe the creep behavior using strain rate formu-
lation in a creep equation like:
𝜀𝜀̇𝑐𝑐𝑐𝑐 = 𝑓𝑓(𝜎𝜎, 𝑡𝑡, 𝜀𝜀, 𝑇𝑇) (1)

320
 Kang Shen, Wilhelm J.H. Rust

It shows that the creep strain rate depends on stress, time (or creep strain) and temperature. The
dependencies of time and strain are normally not modelled within the same creep equation. As
example, a creep equation of direct time dependency (also called time hardening) is defined as
𝑐𝑐4
𝜀𝜀̇𝑐𝑐𝑐𝑐 = 𝐶𝐶1 𝜎𝜎 𝑐𝑐2 𝑡𝑡 𝑐𝑐3 ⋅ 𝑒𝑒 − 𝑇𝑇 (2)

whereas an example for indirect time dependency (strain-hardening), which show better accu-
racy if the stress significantly changes (or creep process is interrupted) during the simulation,
reads
𝑐𝑐4
𝜀𝜀̇𝑐𝑐𝑐𝑐 = 𝐶𝐶1 𝜎𝜎 𝑐𝑐2 (𝜀𝜀 𝑐𝑐𝑐𝑐 )𝑐𝑐3 ⋅ 𝑒𝑒 − 𝑇𝑇 (3)

where the 𝐶𝐶1 … 𝐶𝐶4 are the material parameter. These parameters are fitted to the creep test data
for a number of stress, time (or creep strain) and temperature points. The creep strain rate for a
single curve may be accurate. However, the parameter identification goes easily wrong consid-
ering multiple stress levels at different time point. The reason is that the stress, time and creep
strain have a complex relation. Furthermore, defining the material parameter for indirect time
dependency is harder than for the direct way, since the creep strain rate depends on the creep
strain which makes the solution of the differential equation more difficult.

4 SURFACE CONSTRUCTION
In our new creep model, we need a three-dimensional surface to describe the creep strain
depending on time and stress.1 The new creep model uses three types of creep test data for the
users’ convenience. The creep test data can be the creep-time curve, isochronous curve and
creep modulus curve.
The creep curve uses stress as curve parameter and describes the increase of creep strain
with time. The isochronous curve connects stress-strain points at the same time level and is not
a directly measured curve. These curves describe the strain-stress-time relation. Since the creep
strain is our primary variable, the surface should be
𝜀𝜀 𝑐𝑐𝑐𝑐 = 𝑓𝑓𝑐𝑐𝑐𝑐 (𝜎𝜎, 𝑡𝑡) (4)
To create the surface with the given curves as tabulated data, we need to construct the curves
in both 𝜎𝜎- and 𝑡𝑡-directions first. The curve requires to smoothly cross all the data points in one
direction due to accuracy and differentiability. Thus, the cubic spline is used to create the curve.
The advantage to build the cubic spline is that the spline exactly meets selected data points,
avoid an unnecessary oscillation and provide 𝐶𝐶 2 -continuity. The spline for creep strain and time
uses the logarithmic timeline due to rapid change of the creep strain at the beginning.
The surface is based on cubic splines as boundary curves. It should fit the splines and be
smooth perpendicular to the edges. For this purpose, we use the bicubical Coons patches. The
Coons patches use the boundary curves to generate surfaces [12, 13]. A bicubical Coons patch
combines four edges from splines and the surface interpolation with the cubic Hermite Interpo-
lation. We define the Hermite functions with cubical Bézier form as:

The temperature effect is discussed in section 6.

1

321
 Kang Shen, Wilhelm J.H. Rust

𝐵𝐵03 (𝜎𝜎) = (1 − 𝜎𝜎)3 

𝐵𝐵13 (𝜎𝜎) = 3𝜎𝜎(1 − 𝜎𝜎)2 
(5)
𝐵𝐵23 (𝜎𝜎) = 3𝜎𝜎2 (1 − 𝜎𝜎)
𝐵𝐵33 (𝜎𝜎) = 𝜎𝜎3
So that Hermite function for 𝜎𝜎 ∈ [𝑎𝑎, 𝑏𝑏] is:
𝐻𝐻03 (𝜎𝜎) = 𝐵𝐵03 (𝜎𝜎) + 𝐵𝐵13 (𝜎𝜎)
1
𝐻𝐻13 (𝜎𝜎) = (𝑏𝑏 − 𝑎𝑎)𝐵𝐵13 (𝜎𝜎)
3 (6)
1
𝜎𝜎) = − (𝑏𝑏 − 𝑎𝑎)𝐵𝐵23 (𝜎𝜎)
𝐻𝐻23 (
3
𝐻𝐻33 (𝜎𝜎) = 𝐵𝐵23 (𝜎𝜎) + 𝐵𝐵33 (𝜎𝜎)
From the cubical spline, the positional data is available for Coons patch, the four splines are:
𝑓𝑓(𝑎𝑎, 𝑡𝑡), 𝑓𝑓(𝑏𝑏, 𝑡𝑡), 𝑓𝑓(𝜎𝜎, 𝑐𝑐), 𝑓𝑓(𝜎𝜎, 𝑑𝑑) (7)
which 𝜎𝜎 ∈ [𝑎𝑎, 𝑏𝑏], 𝑡𝑡 ∈ [𝑐𝑐, 𝑑𝑑]. For purpose of continuity, first derivative information for both 𝜎𝜎-
and t-direction is desired but not given. The two patches, which share the same edge in 𝜎𝜎- or t-
direction, must have the same derivative in t- resp. 𝜎𝜎-direction of the edge. Therefore, the de-
rivative is defined through a linear interpolation from two ends of the edge. Hence, there are
four derivatives for each boundary:
𝜕𝜕𝜕𝜕(𝑎𝑎, 𝑡𝑡) 𝜕𝜕𝜕𝜕(𝑏𝑏, 𝑡𝑡) 𝜕𝜕𝜕𝜕(𝜎𝜎, 𝑐𝑐) 𝜕𝜕𝜕𝜕(𝜎𝜎, 𝑑𝑑) (8)
, , ,
𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕
Through four boundary curves and their derivatives two ruled surfaces are defined:
𝜕𝜕𝜕𝜕(𝑎𝑎, 𝑡𝑡) 𝜕𝜕𝜕𝜕(𝑏𝑏, 𝑡𝑡)
ℎ𝑐𝑐(𝜎𝜎, 𝑡𝑡) = 𝐻𝐻03 (𝜎𝜎) ⋅ 𝑓𝑓(𝑎𝑎, 𝑡𝑡) + 𝐻𝐻13 (𝜎𝜎) ⋅ + 𝐻𝐻23 (𝜎𝜎) ⋅ + 𝐻𝐻33 (𝜎𝜎) ⋅ 𝑓𝑓(𝑏𝑏, 𝑡𝑡)
𝜕𝜕𝜎𝜎 𝜕𝜕𝜎𝜎 (9)
𝜕𝜕𝜕𝜕(𝜎𝜎, 𝑐𝑐) 𝜕𝜕𝜕𝜕(𝜎𝜎, 𝑑𝑑)
ℎ𝑑𝑑(𝜎𝜎, 𝑡𝑡) = 𝐻𝐻03 (𝑡𝑡) ⋅ 𝑓𝑓(𝜎𝜎, 𝑐𝑐) + 𝐻𝐻13 (𝑡𝑡) ⋅ + 𝐻𝐻23 (𝑦𝑦) ⋅ + 𝐻𝐻33 (𝑡𝑡) ⋅ 𝑓𝑓(𝜎𝜎, 𝑑𝑑)
𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕
The interpolated surface ℎ𝑐𝑐(𝜎𝜎, 𝑡𝑡) is ruled by the splines 𝑓𝑓(𝑎𝑎, 𝑡𝑡), 𝑓𝑓(𝑏𝑏, 𝑡𝑡) in t-direction for 𝜎𝜎 ∈
[𝑎𝑎, 𝑏𝑏] and ℎ𝑑𝑑(𝜎𝜎, 𝑡𝑡) is ruled by the splines 𝑓𝑓(𝜎𝜎, 𝑐𝑐), 𝑓𝑓(𝜎𝜎, 𝑑𝑑) in 𝜎𝜎-direction for 𝑡𝑡 ∈ [𝑐𝑐, 𝑑𝑑].Thus, we
need ℎ𝑐𝑐(𝜎𝜎, 𝑡𝑡) to fix its course in t-direction and change its course in 𝜎𝜎-direction like ℎ𝑑𝑑(𝜎𝜎, 𝑡𝑡).
In this case, a new surface is used to achieve this goal. We define a surface with the corner data
for the interpolation:
𝜕𝜕𝜕𝜕(𝑎𝑎, 𝑐𝑐)𝜕𝜕𝜕𝜕(𝑎𝑎, 𝑑𝑑)
𝑓𝑓(𝑎𝑎, 𝑐𝑐) 𝑓𝑓(𝑎𝑎, 𝑑𝑑)
𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕
𝜕𝜕𝜕𝜕(𝑎𝑎, 𝑐𝑐) 𝜕𝜕𝜕𝜕(𝑎𝑎, 𝑑𝑑)
0 0
ℎ𝑐𝑐𝑐𝑐(𝜎𝜎, 𝑡𝑡) = (𝐻𝐻03 (𝜎𝜎) 𝐻𝐻13 (𝜎𝜎)𝐻𝐻23 (𝜎𝜎) 𝐻𝐻33 (𝜎𝜎)) ⋅ 𝜕𝜕𝜎𝜎 𝜕𝜕𝜎𝜎
𝜕𝜕𝜕𝜕(𝑏𝑏, 𝑐𝑐) 𝜕𝜕𝜕𝜕(𝑏𝑏, 𝑑𝑑)
0 0
𝜕𝜕𝜎𝜎 𝜕𝜕𝜎𝜎
𝜕𝜕𝜕𝜕(𝑏𝑏, 𝑐𝑐)𝜕𝜕𝜕𝜕(𝑏𝑏, 𝑑𝑑) (10)
( 𝑓𝑓(𝑏𝑏, 𝑐𝑐) 𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕
𝑓𝑓(𝑏𝑏, 𝑑𝑑) )
𝐻𝐻03 (𝑡𝑡)
𝐻𝐻13 (𝑡𝑡)
⋅
𝐻𝐻23 (𝑡𝑡)
3
(𝐻𝐻3 (𝑡𝑡))

The bicubical Coons Patch is defined as:

ℎ𝑔𝑔(𝜎𝜎, 𝑡𝑡) = ℎ𝑐𝑐(𝜎𝜎, 𝑡𝑡) + ℎ𝑑𝑑(𝜎𝜎, 𝑡𝑡) − ℎ𝑐𝑐𝑐𝑐(𝜎𝜎, 𝑡𝑡) (11)

322
 Kang Shen, Wilhelm J.H. Rust

The surface we construct for isochronous curves with the Coons patches is shown in Figure
1.
0.07

0.06

0.05
creep strain

0.04

0.03

0.02

0.01

25
20
15
0 10 stress [MPa]
1
2 5
3
lg. time [h] 4
5 0

(a) T = 23°C, t = 1~105 h, σeqv = 2~20MPa

0.07 0.07

0.06 0.06

0.05 0.05
creep strain
creep strain

0.04 0.04

0.03 0.03

0.02 0.02

0.01 0.01

0 0

25 25
20 20
15 15
0 10 stress [MPa] 0 10 stress [MPa]
1 1
2 2 5
3 5 3
4 lg. time [h] 4
lg. time [h] 5 0 5 0

(b) T = 60°C, t = 1~105 h, σeqv = 1~100MPa (c)T = 100°C, t = 1 − 104 h, σeqv = 1~10MPa
Figure 1 creep strain with logarithmic time and loads for Ultradur B 2550

The entire surface consists of multiple Coons patches. The cross points in the surface are the
data points from the database. With this surface, the creep strain can be interpolated for one
specific couple of time and stress.

5 NEW CREEP MODEL WITH CREEP STRAIN FORMULATION

In our new creep model, we use the incremental creep strain formulation instead of rate
formulation. Since the creep strain 𝜀𝜀 𝑐𝑐𝑐𝑐 → 0 at the time 𝑡𝑡 → 0, the creep process can start from
time 𝑡𝑡 = 0 without the numerical problem even if the strain rate tends to infinity in a rate for-
mulation. We create a surface from the creep curve, isochronous curve as a creep strain function
of time and stress. Therefore, the incremental creep strain formulation for direct time depend-
ency is
Δ𝜀𝜀 𝑐𝑐𝑐𝑐 = 𝑓𝑓𝑐𝑐𝑐𝑐 (𝜎𝜎, 𝑡𝑡 + Δ𝑡𝑡) − 𝑓𝑓𝑐𝑐𝑐𝑐 (𝜎𝜎, 𝑡𝑡) (12)

Since the surface of creep strain is defined as a three-dimensional function 𝑓𝑓𝑐𝑐𝑐𝑐 , the creep
strain increment is interpolated using the stress 𝜎𝜎, the time 𝑡𝑡 and the time increment Δ𝑡𝑡. The
term 𝑓𝑓𝑐𝑐𝑐𝑐 (𝜎𝜎, 𝑡𝑡) is equal to the creep strain 𝜀𝜀0𝑐𝑐𝑐𝑐 from the last converged state.
To model indirect time dependency, we must determine the point of the surface where the
same creep strain 𝜀𝜀0𝑐𝑐𝑐𝑐 is valid, but now for the actual stress 𝜎𝜎𝑖𝑖+1 . Since the data form triples of
strain, stress and time, the time is the only differing quantity and thus is no longer the real time.
Hence, the time of this point is called pseudo-time 𝜉𝜉 (at point a in Figure 2). The surface 𝑓𝑓𝑐𝑐𝑐𝑐

323
 Kang Shen, Wilhelm J.H. Rust

describes the creep strain depending on time and stress. The inverse function, time as a function
of stress and strain is not given. To solve for the pseudo-time 𝜉𝜉, we use a Newton-Raphson
scheme for the equation
ℎ(𝜎𝜎𝑖𝑖+1 , 𝜉𝜉) = 𝑓𝑓𝑐𝑐𝑐𝑐 (𝜎𝜎𝑖𝑖+1 , 𝜉𝜉) − 𝜀𝜀0𝑐𝑐𝑐𝑐 = 0 (13)

Then we obtain the creep strain 𝜀𝜀 𝑐𝑐𝑐𝑐 after time increment Δ𝑡𝑡 at point b in Figure 2. The dif-
ference of point a and b is the creep increment Δ𝜀𝜀 𝑐𝑐𝑐𝑐 .2
Δ𝜀𝜀 𝑐𝑐𝑐𝑐 (𝜎𝜎𝑖𝑖+1 , 𝜉𝜉 + Δ𝑡𝑡) = 𝑓𝑓𝑐𝑐𝑐𝑐 (𝜎𝜎𝑖𝑖+1 , 𝜉𝜉 + Δ𝑡𝑡) − 𝜀𝜀0𝑐𝑐𝑐𝑐 (14)

Figure 2 creep curve with search process

Alternatively we can also create a surface from the creep modulus curve which uses the creep
modulus depending on time and stress. In this case, the creep modulus is depending variable,
so the surface is
𝐸𝐸 𝑐𝑐𝑐𝑐 = 𝑓𝑓𝑐𝑐𝑐𝑐 (𝜎𝜎, 𝑡𝑡) (15)
The creep modulus is defined as the quotient of stress and total strain. The pseudo-time now
is the time point where the same creep modulus applies as in the converged state but for the
new stress. It is determined by a Newton-Raphson scheme for
ℎ(𝜎𝜎𝑖𝑖+1 , 𝜉𝜉) = 𝑓𝑓𝑐𝑐𝑐𝑐 (𝜎𝜎𝑖𝑖+1 , 𝜉𝜉) − 𝐸𝐸0𝑐𝑐𝑐𝑐 = 0 (16)

where 𝐸𝐸0𝑐𝑐𝑐𝑐 is the creep modulus from last converged state. The 𝐸𝐸0𝑐𝑐𝑐𝑐 is calculated using the creep
strain as:
𝜎𝜎𝑖𝑖 (17)
𝐸𝐸0𝑐𝑐𝑐𝑐 =
𝜀𝜀𝑐𝑐𝑐𝑐 (𝑡𝑡𝑖𝑖 ) + 𝜀𝜀𝑒𝑒𝑒𝑒
As a result, the creep modulus for the time 𝜉𝜉 + Δ𝑡𝑡 is
𝐸𝐸 𝑐𝑐𝑐𝑐 = 𝑓𝑓𝑐𝑐𝑐𝑐 (𝜎𝜎𝑖𝑖+1 , 𝜉𝜉 + Δ𝑡𝑡) (18)

2
Further discussion for pseudo-time and stress out of the surface, see ([14])

324
 Kang Shen, Wilhelm J.H. Rust

To use the same local iteration of creep strain formulation (Figure 3), we convert creep modulus
to creep strain
𝜎𝜎𝑖𝑖+1 𝜎𝜎𝑖𝑖+1 (19)
∆𝜀𝜀𝑐𝑐𝑐𝑐 (𝜎𝜎𝑖𝑖+1 , 𝑡𝑡 + Δ𝑡𝑡) = − − 𝜀𝜀0𝑐𝑐𝑐𝑐
𝐸𝐸𝑐𝑐𝑐𝑐 (𝜎𝜎𝑖𝑖+1 , 𝜉𝜉 + Δ𝑡𝑡) 𝐸𝐸
where the first term is the new total strain and the second one the new elastic strain. Its partial
derivative with respect to 𝜎𝜎 is
𝜕𝜕∆𝜀𝜀𝑐𝑐𝑐𝑐 (𝜎𝜎𝑖𝑖+1 , 𝑡𝑡 + Δ𝑡𝑡) 𝜕𝜕 𝜎𝜎𝑖𝑖+1 1 1 𝜎𝜎 𝜕𝜕𝐸𝐸𝑐𝑐𝑐𝑐 1 (20)
= ( )− = − −
𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕 𝐸𝐸𝑐𝑐𝑐𝑐 (𝜎𝜎𝑖𝑖+1 , 𝜉𝜉 + Δ𝑡𝑡) 𝐸𝐸 𝐸𝐸𝑐𝑐𝑐𝑐 𝐸𝐸𝑐𝑐𝑐𝑐 2 𝜕𝜕𝜕𝜕 𝐸𝐸

and with respect to 𝑡𝑡 :

𝜕𝜕∆𝜀𝜀𝑐𝑐𝑐𝑐 (𝜎𝜎𝑒𝑒𝑒𝑒𝑒𝑒,𝑖𝑖+1 , 𝑡𝑡 + Δ𝑡𝑡) 𝜎𝜎 𝜕𝜕𝐸𝐸𝑐𝑐𝑐𝑐 (21)
=−
𝜕𝜕𝜕𝜕 𝐸𝐸𝑐𝑐𝑐𝑐 2 𝜕𝜕𝜕𝜕
Thus, all three types of test data curves are united by the incremental creep strain formulation.
With the creep strain increment, the three-dimensional creep state is defined
𝜕𝜕𝜕𝜕 (22)
Δ𝜺𝜺𝑐𝑐𝑐𝑐 = Δ𝜀𝜀 𝑐𝑐𝑐𝑐 ⋅
𝜕𝜕𝝈𝝈
where 𝑄𝑄 is plastic potential. This equation is adopted from the flow rule of plastic material,
where the plastic multiplier is replaced by the creep strain increment in this case. Like in plas-
tics, we use the flow rule associated with the yield condition 𝐹𝐹 after von Mises. There is no
threshold like the yield stress, thus creep is always present if there is non-zero equivalent stress.
After the Δ𝜺𝜺𝑐𝑐𝑐𝑐 is determined, the current stress state is
𝝈𝝈 = 𝑬𝑬 (𝜺𝜺𝑡𝑡𝑡𝑡𝑡𝑡 − 𝜺𝜺𝑐𝑐𝑐𝑐 (t i ) − Δ𝜺𝜺𝑐𝑐𝑐𝑐 (𝜉𝜉 + Δ𝑡𝑡, 𝜎𝜎𝑒𝑒𝑒𝑒𝑒𝑒 (𝝈𝝈))) (23)

where 𝑬𝑬 is the elasticity matrix, 𝜺𝜺𝑡𝑡𝑡𝑡𝑡𝑡 is total strain for this step. The stress state changes with
the creep increment updates. The equivalent stress is defined after von Mises. In equation (23),
stress tensor 𝝈𝝈 is defined implicitly. Therefore, a local iteration with Newton-Raphson method
should take place
𝑔𝑔 = Δ𝜀𝜀 𝑐𝑐𝑐𝑐 − Δ𝜀𝜀 𝑐𝑐𝑐𝑐 (𝜎𝜎𝑒𝑒𝑒𝑒𝑒𝑒 (𝝈𝝈), 𝜉𝜉 + Δ𝑡𝑡) − 𝜀𝜀𝑒𝑒𝑒𝑒𝑒𝑒
𝑐𝑐𝑐𝑐 (𝑡𝑡 )
𝑖𝑖 = 0 (24)

The derivative of 𝑔𝑔 is
𝜕𝜕𝜕𝜕 𝜕𝜕Δ𝜀𝜀 𝑐𝑐𝑐𝑐 𝜕𝜕𝜕𝜕 𝑇𝑇 𝜕𝜕𝜕𝜕 𝜕𝜕Δ𝜀𝜀 𝑐𝑐𝑐𝑐 (25)
= 1 + ( ) 𝑬𝑬 −
𝜕𝜕Δ𝜀𝜀 𝑐𝑐𝑐𝑐 𝜕𝜕𝜎𝜎𝑒𝑒𝑒𝑒𝑒𝑒 𝜕𝜕𝝈𝝈 𝜕𝜕𝝈𝝈 𝜕𝜕𝜀𝜀 𝑐𝑐𝑐𝑐

For each iteration, the creep increment and its derivative is determined using the surface of
Coons patches. The process for the creep increment calculation is shown in Figure 3

325
 Kang Shen, Wilhelm J.H. Rust

Figure 3 flow chart for local iteration

Once the creep increment is determined and the local iteration is converged, the consistent
tangent is requested from the global iteration. The total differential of the stress is:
𝜕𝜕𝜕𝜕 𝜕𝜕 2 𝑄𝑄 (26)
𝑑𝑑𝝈𝝈 = 𝑬𝑬 (𝑑𝑑𝜺𝜺𝑡𝑡𝑡𝑡𝑡𝑡 − dΔ𝜀𝜀 𝑐𝑐𝑐𝑐 − Δ𝜀𝜀 𝑐𝑐𝑐𝑐 2 𝑑𝑑𝝈𝝈)
𝜕𝜕𝝈𝝈 𝜕𝜕𝝈𝝈

The total differential of the creep increment is:

𝜕𝜕Δ𝜀𝜀 𝑐𝑐𝑐𝑐 𝜕𝜕𝜕𝜕 𝜕𝜕Δ𝜀𝜀 𝑐𝑐𝑐𝑐 (27)
dΔ𝜀𝜀 𝑐𝑐𝑐𝑐 = 𝑑𝑑𝝈𝝈 + dΔ𝜀𝜀 𝑐𝑐𝑐𝑐
𝜕𝜕𝜎𝜎𝑒𝑒𝑒𝑒𝑒𝑒 𝜕𝜕𝝈𝝈 𝜕𝜕𝜀𝜀 𝑐𝑐𝑐𝑐

Together with equation (26) and (27) a linear system of equations of order n+1 with n col-
umns at the right hand side is obtained:
𝜕𝜕 2 𝑄𝑄 𝜕𝜕𝜕𝜕
𝟏𝟏 + Δ𝜀𝜀 𝑐𝑐𝑐𝑐 𝑬𝑬 𝑬𝑬
𝜕𝜕𝝈𝝈2 𝜕𝜕𝝈𝝈 𝑑𝑑𝝈𝝈 𝑬𝑬
( ) = ( ) 𝑑𝑑𝜺𝜺𝑡𝑡𝑡𝑡𝑡𝑡 (28)
𝜕𝜕Δ𝜀𝜀 𝑐𝑐𝑐𝑐
𝜕𝜕𝜕𝜕 𝑇𝑇 𝜕𝜕Δ𝜀𝜀 𝑐𝑐𝑐𝑐 dΔ𝜀𝜀 𝑐𝑐𝑐𝑐 0
( ) 𝑐𝑐𝑐𝑐
+1
( 𝜕𝜕𝜎𝜎𝑒𝑒𝑒𝑒𝑒𝑒 𝜕𝜕𝝈𝝈 𝜕𝜕𝜀𝜀 )
where n is the number of strain components. The first n rows of the solution form the con-
𝑑𝑑𝝈𝝈
sistent tangent 𝑑𝑑𝜺𝜺𝑡𝑡𝑡𝑡𝑡𝑡 .

6 TEMPERATURE EFFECT
In conventional creep models the Arrhenius equation is often used to characterize the temper-
ature effect.
𝑐𝑐
𝜀𝜀 𝑐𝑐𝑐𝑐̇ = 𝑓𝑓(𝜎𝜎, 𝑡𝑡, 𝜀𝜀) ⋅ 𝑒𝑒 −𝑇𝑇 (29)

The constant c is the quotient of activation energy and universal gas constant R. However, the
constant c is not a measurable parameter during a creep test. The normal way to determine this
parameter is using a curve fitting or solving the equation for selected data points. For the Ma-
terialPBT-GB30 (Ultradur B 4300 K63), we obtain the creep data like:

3
Creep data from [11], the isochronous curves is interpolated using Coons patches. The original data was meas-
ured form different load classes. In order to explain the temperature effect without other disruption, a constant
load is assumed.

326
 Kang Shen, Wilhelm J.H. Rust

Table 1: Temperature-creep relation for PBT-GB30

𝑇𝑇 [𝐾𝐾] 𝜀𝜀 𝑐𝑐𝑐𝑐 c
296,15 2,26E-04
313,15 2,78E-03 1,37E+04
333,15 5,44E-03 3,51E+03
363,15 6,25E-03 5,61E+02
393,15 8,87E-03 1,67E+03
413,15 1,43E-02 3,89E+03
for σ = 5 MPa, t = 10h. Hence the parameter c can be determined from two data points in the
Table 1:
𝜀𝜀̇ 𝑖𝑖𝑐𝑐𝑐𝑐 (30)
𝑙𝑙𝑙𝑙 𝑐𝑐𝑐𝑐
𝜀𝜀̇ 𝑖𝑖+1
𝑐𝑐𝑖𝑖 = 1 1
−
𝑇𝑇𝑖𝑖+1 𝑇𝑇𝑖𝑖

Using this parameter c for the temperature outside of these two data points causes an error
(Figure 4).
A linear interpolation for parameter c can improve the behavior (Figure 4) but does not solve
the problem that the creep curves for different temperatures can show different shape.
The new model for this creep strain-temperature relation uses the Coons patches (Figure 5).
Once the stress and the time are set, we obtain the creep strain εcr from the surface of Coons
patch as interpolated points. A cubical spline connects interpolated points and creates a creep
strain-temperature function. Thus, the creep strain could be determined from a specific temper-
ature within the spline. This solution creates a smooth creep strain-temperature curve (Figure
4). Since temperature is a given value during the simulation no derivative with respect to it is
requested. Thus, lower order piecewise interpolation is possible. It might be the more accurate
the more temperature points are available from test data. For a smaller number of temperatures
spline interpolation is preferable.
0,016
0,014
0,012
creep strain

0,01
0,008
0,006
0,004
0,002
0
290 310 330 350 370 390 410 430 450
Temperatur [K]
data point c as constant c linear interpolated Coons patch interpolant
Figure 4 Creep strain-temperature relation using interpolation

327
 Kang Shen, Wilhelm J.H. Rust

If the temperature changes during the test, this interpolation takes place in every step. If the
simulation takes place at a specific constant temperature but not matching a test temperature
we first process the creep strain-temperature function for the corner points and create a new
surface for creep strain-stress-time relation at this temperature before the simulation starts.

Figure 5 new model illustration for interpolation of the temperature effect

5 RESULTS
Figure 6 shows the results of this new creep model. To compute these results, a solid element
with 8 nodes is used. The curve “00iso” uses the isochronous curve and the curve“00cm” uses
the creep modulus curve. Both curves start the creep at 𝑡𝑡 = 0 ℎ. They match the data point
exactly. The maximum difference between the two curves amounts to 1.12%.

0,4
Creep Strain[%]

0,3

0,2

0,1

0
0,00 1,00 2,00 3,00 4,00 5,00 6,00 7,00 8,00 9,00 10,00
Time [h]

DP1 DP10 00iso 05iso 00km 05km 05iso direct

Figure 6 creep curve using the new model in compare with data points

The other two results (05iso and 05cm) start the creep calculation after 30min with indirect
time dependency. Until then the creep strain keeps zero. After 30 minutes, they start with the

10

328
 Kang Shen, Wilhelm J.H. Rust

same curve as“00iso” and “00cm”. This result shows that the indirect time dependency work
properly. For the direct time dependency the curve “05iso direct” is considering the creep pro-
cess as already done for 30 minutes, so it starts with the range of the curve “00iso” beginning
at the same time point. Thus, this creep model works as it should be.
We implement this creep model into the commercial software ANSYS. Figure 7 shows two
examples for the use of the creep model. First model is a tensile specimen. One side of the
specimen is fixed, on the other side a tensile load of 10 MPa is applied. As we expect, the creep
strain concentrate on the reduced cross section. The second model is a plate with a hole in it.
We apply the same boundary conditions to this model. In this model, we observe the creep
concentration and the gradient of the creep strain.
These results show that the new creep model works properly with a complex geometry and
multiple elements. It provides also a convergence like other creep models. The huge advantage
of this new creep model is that:
 Only tabulated creep test data are used as input data
 No material parameters, are required, i.e. no curve fitting must take place
 The surface of Coons patches meets every data point.

Figure 7 two examples: a tensile specimen and a plate with hole

REFERENCES
[1] L. C. Brinson and T. S. Gates, “Effects of physical aging on long term creep of polymers
and polymer matrix composites,” International Journal of Solids and Structures, vol. 32,
no. 6-7, pp. 827–846, 1995.
[2] C. Marais and G. Villoutreix, “Analysis and modeling of the creep behavior of the ther-
mostable PMR-15 polyimide,” J. Appl. Polym. Sci., vol. 69, no. 10, pp. 1983–1991,
1998.
[3] M. Ezrin, Plastics failure guide: Cause and prevention, 2nd ed. Munich: Hanser, 2013.
[4] W. Rust, Non-Linear Finite-Elemente Analysis in Structural Mechanics, Cham/Heidel-
berg: Springer, 2015.
[5] M. Stommel, M. Stojek, and W. Korte, FEM zur Berechnung von Kunststoff- und
Elastomerbauteilen. München: Hanser Carl, 2011.
[6] I. SAS IP, Ansys Help, 2016.
[7] F. Vogt, Simulation von Kriechgesetzen, student project, Hochschule Hannover, 2016.

11

329
 Kang Shen, Wilhelm J.H. Rust

[8] M. J. Dropik, D. H. Johnson, and D. E. Roth, “Developing an ANSYS creep model for
polypropylene from experimental data,” in Proceedings of International ANSYS Confer-
ence, 2002.
[9] Y. S. Kim and D. R. Metzger, “A finite element study of friction effect in four-point
bending creep test,” in ASME 2003 Pressure Vessels and Piping Conference, 2003, pp.
99–107.
[10] T. Pulngern et al., “Effect of temperature on mechanical properties and creep responses
for wood/PVC composites,” Construction and Building Materials, vol. 111, pp. 191–198,
2016.
[11] M-Base Engineering + Software GmbH, CAMPUS® Plastics Material Database.
[Online] Available: http://www.campusplastics.com/campus/.
[12] G. Farin, Curves and surfaces for CAGD: A practical guide, 5th ed. San Francisco, Ca-
lif.: Morgan Kaufmann Publ, 2006.
[13] S. A. Coons, “Surfaces for computer-aided design of space forms,” DTIC Document,
1967.
[14] Wilhelm J.H. Rust, Kang Shen, “Ein Kriechmodell für ANSYS unter direkter Verwen-
dung von Versuchsdaten”,Proceedings of 34. CADFEM ANSYS Simulation Conference,
2016.

12

330